Investors are “buying the dream.”

That’s the message from BofA’s Savita Subramanian headed into 2026, when US equities will manage a mere 5% gain, according to an expansive year-ahead outlook which features an overtly cautious take on the market leadership.

“Investors should get ready for an air pocket,” Subramanian warned, of the AI trade, noting that “monetization is to be determined and power is the bottle neck and will take a while to build out.” (What a bummer, right?)

You all know the narrative. At this juncture, it only bears repeating because it’s the most important story on the board which, now that I think about it, is a pretty damn good reason for repetition.

The issue’s captured in the figure below which I’ve run I don’t know how many times by now.

The hyper-scalers can’t fund capex plans with free cash flow anymore — not unless they want to spend eight out of every $10 on the AI buildout. So: Borrowing. Lots of it.

“The supply/demand backdrop for Tech debt just worsened significantly,” Subramanian remarked, noting that YoY, supply rose tenfold in 2025.

Some of you have likely seen the figure below by now, but I figured I’d highlight it just in case.

“Hyper-scalers have been mostly rewarded for increasing spend but capex has shifted from being funded by operating cash flow to debt raises and hyper-scalers’ capital intensity rocketed up from 13% in 2012 to 64% today,” Subramanian said, noting that these companies “are now more capital intensive than the oil majors.”

Is this something to worry about? Well, yes. Probably. And for a lot of different reasons.

For one thing, this is an seismic shift: Most of these companies were traditionally capital-light, and while management’s world-class, the very nature of their businesses and balance sheets have changed virtually overnight. That’s a challenging transition for any C-suite to make in such a compressed time frame.

In addition, I’m still unconvinced — even more so in light of recent events around the Gemini 3 coup — that this tech epoch’s going to break the mold, where that means everything will get bigger and more expensive as opposed to smaller and cheaper. If this time isn’t different in that regard, some (a lot) of current spending will look wasteful in hindsight.

Finally, borrowing costs are a function of supply and demand. The more supply, the higher the cost, even for the best credits. Indeed, tech spreads have now widened to (roughly) match those of the broader IG universe.

“Return on incremental invested capital versus cost of capital is a good guidepost for what multiple to pay for AI stocks,” Subramanian wrote. “The spread today for most hyper-scalers is healthy if not stellar, but returns are likely to narrow and the cost of capital is likely to increase from higher leverage, where today’s spreads and equity risk premia are near record lows.”

The title of the section in her year-ahead dedicated to the hyper-scalers reads, “How does AI break?”